Ask HN: What does your production machine learning pipeline look like?
Technology#Machine Learning Pipelines👥 Community|Analyzed: Jan 3, 2026 06:30•
Published: Mar 8, 2017 16:15
•1 min read
•Hacker NewsAnalysis
The article is a discussion starter on Hacker News, soliciting information about production machine learning pipelines. It presents a specific example using Spark, PMML, Openscoring, and Node.js, highlighting the separation of training and execution. It also raises a question about the challenges of using technologies like TensorFlow where model serialization and deployment are more tightly coupled.
Key Takeaways
- •The article describes a production ML pipeline using Spark for training, PMML for model representation, Openscoring for model serving, and Node.js for a web service wrapper.
- •It highlights the trade-offs of separating training and execution technologies, particularly the limitations of PMML.
- •It raises questions about the model deployment process for technologies like TensorFlow, where model serialization and execution are more integrated.
Reference / Citation
View Original"Model training happened nightly on a Spark cluster... Separating the training technology from the execution technology was nice but the PMML format is limiting..."